GPU benchmarks

[DPGrafik]

Sure thing - The eGPU will be up and running within two weeks, thats when i expect that final parts to arrive. It will be interesting to see both if it works and if it's significantly slower than internal GPU due speed limitations with connection. I'm using thunderbolt 2 to connect, so theoretically it should do 20GB/s but i think i read somewhere that with pci-e over thunderbolt 2 its limited to 5Gb/s, what that means for rendering we'll see once its running .

[savage309]

Sure thing - The eGPU will be up and running within two weeks, thats when i expect that final parts to arrive. It will be interesting to see both if it works and if it's significantly slower than internal GPU due speed limitations with connection. I'm using thunderbolt 2 to connect, so theoretically it should do 20GB/s but i think i read somewhere that with pci-e over thunderbolt 2 its limited to 5Gb/s, what that means for rendering we'll see once its running .

RT GPU can be tuned for that. If you increase ray bundle size / rays per pixel, bigger chunks of work will be given to the GPUs, this will keep them more busy and will reduce the traffic (CPU to GPU and/or between DR nodes). This will reduce the interactivity a bit though. We have that pretty well documented here.

[DPGrafik]

Cool i'll look into it. Can't wait to see how this stack up against the i7 render nodes i have now.

[tricky]

My render dialogue tells me this took 2:43 to render and this is on a GTX980Ti and the latest drivers. Its good, but looking at the list of other cards, shouldn't I be expecting faster times?

I am on 3.40.01 on Max2016. I notice a couple of missing texture maps when I open the scene, but other than that, I just switched to CUDA and clicked render

[savage309]

I think you are not starting the correct test. Do you have log / from where you got the scene ?
Best,
Blago.

[tricky]

I think you are not starting the correct test. Do you have log / from where you got the scene ?
Best,
Blago.

I think I must have the wrong scene. Can you give me a link to download the correct one please?

(EDIT - Ah. I think I've found it)

[tricky]

I've downloaded the scene from the link on the spreadsheet. The render just keeps on going. Should it stop at 2048 paths/pixel? Its currently at 2688 paths/pixel and still going.

[savage309]

I've downloaded the scene from the link on the spreadsheet. The render just keeps on going. Should it stop at 2048 paths/pixel? Its currently at 2688 paths/pixel and still going.

There are two benchmarks. 1.0 and 2.0.
1.0 should run in less than 2 minutes, the other one in less than 20 minutes with your GPU.
There are links for downloading both of them in the spreadsheet. Keep in mind that between different V-Ray versions the time may vary a lot. For the same amount of pixels it gives the different noise level, etc.
V-Ray 3.30 and later can do a bit more paths per pixel than the specified in the settings, so this is normal as well.

[werticus]

Still no one with a 1080? Common :P

[FSGFX]

Vlado did some tests on google+
https://plus.google.com/+VladimirKoy...ts/R1XPacoWvB4

[Morne]

Would love to see how AMD's 480 does with this...

[Sushidelic]

I can confirm that 1080's performance is pretty much the same as a TitanX right now. Waiting and hoping for Cuda8 to hit the streets in new drivers, squeezing the last out of the Pascals. 1080 should be ~150% of a TitanX if you look at raw Teraflops only.

[savage309]

1080 is a very good GPU, but it is a replacement for 980, not for the Titan X. It is quite faster than the 980 and has more RAM, and fast as Titan X, which is great.
With the new drivers I hope there to be some improvement as well, but I don't think it will get much faster than it is today.
Btw, teraflops should be compared only between the same architectures, otherwise they don't mean too much.

Best,
Blago.

[Sushidelic]

That's kinda sad news, was still hoping that the raw power was yet to be unleashed... Anyway, it's blazing fast, costing less and - more important - way less TDP than a TitanX. 8GB vs. 12GB is the big drawback, let's hope that there's gonna be a 16GB version of a Pascal based Titan, as it looks like there's room for twice the RAM installed on the PCB of the 1080 already. Right now, rumors are, those slots are only reserved for their Quadro lineup, but who knows. Else it might take till late 2017, that real new gaming devices will touch base, so investing in a bunch of 1080ies seemed not too wrong.
Could well be that I show up a little bit more on the forum in the near future, as I used to do in the very early days of V-Ray CPU - imho things are changing again, like going from 32bit OS to 64bit, or like the change from 16bit int to EXR float images. GPU is massive, and it will change the market. 20x less energy is only one small side of things. In ten years, rendering on CPU will be similar like still rendering out to TGA. CPU is dead.

[Nicinus]

In ten years, rendering on CPU will be similar like still rendering out to TGA. CPU is dead.

Problem is, I need to render today...